Deep Learning Based Infrared Image Recognize and Internal Overheating Fault Diagnosis of Gas Insulated Switchgear

Current infrared inspection method could not identify internal connector temperature rise on different components of gas insulated switchgear (GIS). In this paper, a deep learning based infrared image recognize and internal overheating fault diagnosis method of GIS is purposed. Firstly, electromagnetic-fluid-thermal coupled numerical calculation are carried out to obtain relationship between surface and internal connector temperature rises, and temperature rise of internal connector is obtained from this relationship through calculated surface temperature. Secondly, various components of 110kV/220kV GIS which include are labeled from 700 field infrared images. A MASK R-CNN model is adopted to recognize different components of GIS. Finally, surface temperature of different GIS components is calculated by grey level values of segmented infrared images. The GIS component recognition algorithm and the internal connector temperature inversion algorithm are integrated into the software developed based on PyQt. On field application, good application effects can be achieved. Field applications show that the average recognition and diagnosis time for each infrared picture is about 4 seconds, which meets the real-time application requirements in the actual scene of power inspection.